1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
// * This file is part of the uutils coreutils package.
// *
// * For the full copyright and license information, please view the LICENSE
// * file that was distributed with this source code.
// spell-checker:ignore (ToDO) tailable untailable stdlib kqueue Uncategorized unwatch
use crate::args::{FollowMode, Settings};
use crate::follow::files::{FileHandling, PathData};
use crate::paths::{Input, InputKind, MetadataExtTail, PathExtTail};
use crate::{platform, text};
use notify::{RecommendedWatcher, RecursiveMode, Watcher, WatcherKind};
use std::io::BufRead;
use std::path::{Path, PathBuf};
use std::sync::mpsc::{self, channel, Receiver};
use uucore::display::Quotable;
use uucore::error::{set_exit_code, UResult, USimpleError};
use uucore::show_error;
pub struct WatcherRx {
watcher: Box<dyn Watcher>,
receiver: Receiver<Result<notify::Event, notify::Error>>,
}
impl WatcherRx {
fn new(
watcher: Box<dyn Watcher>,
receiver: Receiver<Result<notify::Event, notify::Error>>,
) -> Self {
Self { watcher, receiver }
}
/// Wrapper for `notify::Watcher::watch` to also add the parent directory of `path` if necessary.
fn watch_with_parent(&mut self, path: &Path) -> UResult<()> {
let mut path = path.to_owned();
#[cfg(target_os = "linux")]
if path.is_file() {
/*
NOTE: Using the parent directory instead of the file is a workaround.
This workaround follows the recommendation of the notify crate authors:
> On some platforms, if the `path` is renamed or removed while being watched, behavior may
> be unexpected. See discussions in [#165] and [#166]. If less surprising behavior is wanted
> one may non-recursively watch the _parent_ directory as well and manage related events.
NOTE: Adding both: file and parent results in duplicate/wrong events.
Tested for notify::InotifyWatcher and for notify::PollWatcher.
*/
if let Some(parent) = path.parent() {
if parent.is_dir() {
path = parent.to_owned();
} else {
path = PathBuf::from(".");
}
} else {
return Err(USimpleError::new(
1,
format!("cannot watch parent directory of {}", path.display()),
));
};
}
if path.is_relative() {
path = path.canonicalize()?;
}
// for syscalls: 2x "inotify_add_watch" ("filename" and ".") and 1x "inotify_rm_watch"
self.watch(&path, RecursiveMode::NonRecursive)?;
Ok(())
}
fn watch(&mut self, path: &Path, mode: RecursiveMode) -> UResult<()> {
self.watcher
.watch(path, mode)
.map_err(|err| USimpleError::new(1, err.to_string()))
}
fn unwatch(&mut self, path: &Path) -> UResult<()> {
self.watcher
.unwatch(path)
.map_err(|err| USimpleError::new(1, err.to_string()))
}
}
pub struct Observer {
/// Whether --retry was given on the command line
pub retry: bool,
/// The [`FollowMode`]
pub follow: Option<FollowMode>,
/// Indicates whether to use the fallback `polling` method instead of the
/// platform specific event driven method. Since `use_polling` is subject to
/// change during runtime it is moved out of [`Settings`].
pub use_polling: bool,
pub watcher_rx: Option<WatcherRx>,
pub orphans: Vec<PathBuf>,
pub files: FileHandling,
pub pid: platform::Pid,
}
impl Observer {
pub fn new(
retry: bool,
follow: Option<FollowMode>,
use_polling: bool,
files: FileHandling,
pid: platform::Pid,
) -> Self {
let pid = if platform::supports_pid_checks(pid) {
pid
} else {
0
};
Self {
retry,
follow,
use_polling,
watcher_rx: None,
orphans: Vec::new(),
files,
pid,
}
}
pub fn from(settings: &Settings) -> Self {
Self::new(
settings.retry,
settings.follow,
settings.use_polling,
FileHandling::from(settings),
settings.pid,
)
}
pub fn add_path(
&mut self,
path: &Path,
display_name: &str,
reader: Option<Box<dyn BufRead>>,
update_last: bool,
) -> UResult<()> {
if self.follow.is_some() {
let path = if path.is_relative() {
std::env::current_dir()?.join(path)
} else {
path.to_owned()
};
let metadata = path.metadata().ok();
self.files.insert(
&path,
PathData::new(reader, metadata, display_name),
update_last,
);
}
Ok(())
}
pub fn add_stdin(
&mut self,
display_name: &str,
reader: Option<Box<dyn BufRead>>,
update_last: bool,
) -> UResult<()> {
if self.follow == Some(FollowMode::Descriptor) {
return self.add_path(
&PathBuf::from(text::DEV_STDIN),
display_name,
reader,
update_last,
);
}
Ok(())
}
pub fn add_bad_path(
&mut self,
path: &Path,
display_name: &str,
update_last: bool,
) -> UResult<()> {
if self.retry && self.follow.is_some() {
return self.add_path(path, display_name, None, update_last);
}
Ok(())
}
pub fn start(&mut self, settings: &Settings) -> UResult<()> {
if settings.follow.is_none() {
return Ok(());
}
let (tx, rx) = channel();
/*
Watcher is implemented per platform using the best implementation available on that
platform. In addition to such event driven implementations, a polling implementation
is also provided that should work on any platform.
Linux / Android: inotify
macOS: FSEvents / kqueue
Windows: ReadDirectoryChangesWatcher
FreeBSD / NetBSD / OpenBSD / DragonflyBSD: kqueue
Fallback: polling every n seconds
NOTE:
We force the use of kqueue with: features=["macos_kqueue"].
On macOS only `kqueue` is suitable for our use case because `FSEvents`
waits for file close util it delivers a modify event. See:
https://github.com/notify-rs/notify/issues/240
*/
let watcher: Box<dyn Watcher>;
let watcher_config = notify::Config::default()
.with_poll_interval(settings.sleep_sec)
/*
NOTE: By enabling compare_contents, performance will be significantly impacted
as all files will need to be read and hashed at each `poll_interval`.
However, this is necessary to pass: "gnu/tests/tail-2/F-vs-rename.sh"
*/
.with_compare_contents(true);
if self.use_polling || RecommendedWatcher::kind() == WatcherKind::PollWatcher {
self.use_polling = true; // We have to use polling because there's no supported backend
watcher = Box::new(notify::PollWatcher::new(tx, watcher_config).unwrap());
} else {
let tx_clone = tx.clone();
match notify::RecommendedWatcher::new(tx, notify::Config::default()) {
Ok(w) => watcher = Box::new(w),
Err(e) if e.to_string().starts_with("Too many open files") => {
/*
NOTE: This ErrorKind is `Uncategorized`, but it is not recommended
to match an error against `Uncategorized`
NOTE: Could be tested with decreasing `max_user_instances`, e.g.:
`sudo sysctl fs.inotify.max_user_instances=64`
*/
show_error!(
"{} cannot be used, reverting to polling: Too many open files",
text::BACKEND
);
set_exit_code(1);
self.use_polling = true;
watcher = Box::new(notify::PollWatcher::new(tx_clone, watcher_config).unwrap());
}
Err(e) => return Err(USimpleError::new(1, e.to_string())),
};
}
self.watcher_rx = Some(WatcherRx::new(watcher, rx));
self.init_files(&settings.inputs)?;
Ok(())
}
pub fn follow_descriptor(&self) -> bool {
self.follow == Some(FollowMode::Descriptor)
}
pub fn follow_name(&self) -> bool {
self.follow == Some(FollowMode::Name)
}
pub fn follow_descriptor_retry(&self) -> bool {
self.follow_descriptor() && self.retry
}
pub fn follow_name_retry(&self) -> bool {
self.follow_name() && self.retry
}
fn init_files(&mut self, inputs: &Vec<Input>) -> UResult<()> {
if let Some(watcher_rx) = &mut self.watcher_rx {
for input in inputs {
match input.kind() {
InputKind::Stdin => continue,
InputKind::File(path) => {
#[cfg(all(unix, not(target_os = "linux")))]
if !path.is_file() {
continue;
}
let mut path = path.to_owned();
if path.is_relative() {
path = std::env::current_dir()?.join(path);
}
if path.is_tailable() {
// Add existing regular files to `Watcher` (InotifyWatcher).
watcher_rx.watch_with_parent(&path)?;
} else if !path.is_orphan() {
// If `path` is not a tailable file, add its parent to `Watcher`.
watcher_rx
.watch(path.parent().unwrap(), RecursiveMode::NonRecursive)?;
} else {
// If there is no parent, add `path` to `orphans`.
self.orphans.push(path);
}
}
}
}
}
Ok(())
}
#[allow(clippy::cognitive_complexity)]
fn handle_event(
&mut self,
event: ¬ify::Event,
settings: &Settings,
) -> UResult<Vec<PathBuf>> {
use notify::event::*;
let event_path = event.paths.first().unwrap();
let mut paths: Vec<PathBuf> = vec![];
let display_name = self.files.get(event_path).display_name.clone();
match event.kind {
EventKind::Modify(ModifyKind::Metadata(MetadataKind::Any | MetadataKind::WriteTime))
// | EventKind::Access(AccessKind::Close(AccessMode::Write))
| EventKind::Create(CreateKind::File | CreateKind::Folder | CreateKind::Any)
| EventKind::Modify(ModifyKind::Data(DataChange::Any))
| EventKind::Modify(ModifyKind::Name(RenameMode::To)) => {
if let Ok(new_md) = event_path.metadata() {
let is_tailable = new_md.is_tailable();
let pd = self.files.get(event_path);
if let Some(old_md) = &pd.metadata {
if is_tailable {
// We resume tracking from the start of the file,
// assuming it has been truncated to 0. This mimics GNU's `tail`
// behavior and is the usual truncation operation for log self.files.
if !old_md.is_tailable() {
show_error!( "{} has become accessible", display_name.quote());
self.files.update_reader(event_path)?;
} else if pd.reader.is_none() {
show_error!( "{} has appeared; following new file", display_name.quote());
self.files.update_reader(event_path)?;
} else if event.kind == EventKind::Modify(ModifyKind::Name(RenameMode::To))
|| (self.use_polling
&& !old_md.file_id_eq(&new_md)) {
show_error!( "{} has been replaced; following new file", display_name.quote());
self.files.update_reader(event_path)?;
} else if old_md.got_truncated(&new_md)? {
show_error!("{}: file truncated", display_name);
self.files.update_reader(event_path)?;
}
paths.push(event_path.to_owned());
} else if !is_tailable && old_md.is_tailable() {
if pd.reader.is_some() {
self.files.reset_reader(event_path);
} else {
show_error!(
"{} has been replaced with an untailable file",
display_name.quote()
);
}
}
} else if is_tailable {
show_error!( "{} has appeared; following new file", display_name.quote());
self.files.update_reader(event_path)?;
paths.push(event_path.to_owned());
} else if settings.retry {
if self.follow_descriptor() {
show_error!(
"{} has been replaced with an untailable file; giving up on this name",
display_name.quote()
);
let _ = self.watcher_rx.as_mut().unwrap().watcher.unwatch(event_path);
self.files.remove(event_path);
if self.files.no_files_remaining(settings) {
return Err(USimpleError::new(1, text::NO_FILES_REMAINING));
}
} else {
show_error!(
"{} has been replaced with an untailable file",
display_name.quote()
);
}
}
self.files.update_metadata(event_path, Some(new_md));
}
}
EventKind::Remove(RemoveKind::File | RemoveKind::Any)
// | EventKind::Modify(ModifyKind::Name(RenameMode::Any))
| EventKind::Modify(ModifyKind::Name(RenameMode::From)) => {
if self.follow_name() {
if settings.retry {
if let Some(old_md) = self.files.get_mut_metadata(event_path) {
if old_md.is_tailable() && self.files.get(event_path).reader.is_some() {
show_error!(
"{} {}: {}",
display_name.quote(),
text::BECOME_INACCESSIBLE,
text::NO_SUCH_FILE
);
}
}
if event_path.is_orphan() && !self.orphans.contains(event_path) {
show_error!("directory containing watched file was removed");
show_error!(
"{} cannot be used, reverting to polling",
text::BACKEND
);
self.orphans.push(event_path.to_owned());
let _ = self.watcher_rx.as_mut().unwrap().unwatch(event_path);
}
} else {
show_error!("{}: {}", display_name, text::NO_SUCH_FILE);
if !self.files.files_remaining() && self.use_polling {
// NOTE: GNU's tail exits here for `---disable-inotify`
return Err(USimpleError::new(1, text::NO_FILES_REMAINING));
}
}
self.files.reset_reader(event_path);
} else if self.follow_descriptor_retry() {
// --retry only effective for the initial open
let _ = self.watcher_rx.as_mut().unwrap().unwatch(event_path);
self.files.remove(event_path);
} else if self.use_polling && event.kind == EventKind::Remove(RemoveKind::Any) {
/*
BUG: The watched file was removed. Since we're using Polling, this
could be a rename. We can't tell because `notify::PollWatcher` doesn't
recognize renames properly.
Ideally we want to call seek to offset 0 on the file handle.
But because we only have access to `PathData::reader` as `BufRead`,
we cannot seek to 0 with `BufReader::seek_relative`.
Also because we don't have the new name, we cannot work around this
by simply reopening the file.
*/
}
}
EventKind::Modify(ModifyKind::Name(RenameMode::Both)) => {
/*
NOTE: For `tail -f a`, keep tracking additions to b after `mv a b`
(gnu/tests/tail-2/descriptor-vs-rename.sh)
NOTE: The File/BufReader doesn't need to be updated.
However, we need to update our `files.map`.
This can only be done for inotify, because this EventKind does not
trigger for the PollWatcher.
BUG: As a result, there's a bug if polling is used:
$ tail -f file_a ---disable-inotify
$ mv file_a file_b
$ echo A >> file_b
$ echo A >> file_a
The last append to file_a is printed, however this shouldn't be because
after the "mv" tail should only follow "file_b".
TODO: [2022-05; jhscheer] add test for this bug
*/
if self.follow_descriptor() {
let new_path = event.paths.last().unwrap();
paths.push(new_path.to_owned());
let new_data = PathData::from_other_with_path(self.files.remove(event_path), new_path);
self.files.insert(
new_path,
new_data,
self.files.get_last().unwrap() == event_path
);
// Unwatch old path and watch new path
let _ = self.watcher_rx.as_mut().unwrap().unwatch(event_path);
self.watcher_rx.as_mut().unwrap().watch_with_parent(new_path)?;
}
}
_ => {}
}
Ok(paths)
}
}
#[allow(clippy::cognitive_complexity)]
pub fn follow(mut observer: Observer, settings: &Settings) -> UResult<()> {
if observer.files.no_files_remaining(settings) && !observer.files.only_stdin_remaining() {
return Err(USimpleError::new(1, text::NO_FILES_REMAINING.to_string()));
}
let mut process = platform::ProcessChecker::new(observer.pid);
let mut _event_counter = 0;
let mut _timeout_counter = 0;
// main follow loop
loop {
let mut _read_some = false;
// If `--pid=p`, tail checks whether process p
// is alive at least every `--sleep-interval=N` seconds
if settings.follow.is_some() && observer.pid != 0 && process.is_dead() {
// p is dead, tail will also terminate
break;
}
// For `-F` we need to poll if an orphan path becomes available during runtime.
// If a path becomes an orphan during runtime, it will be added to orphans.
// To be able to differentiate between the cases of test_retry8 and test_retry9,
// here paths will not be removed from orphans if the path becomes available.
if observer.follow_name_retry() {
for new_path in &observer.orphans {
if new_path.exists() {
let pd = observer.files.get(new_path);
let md = new_path.metadata().unwrap();
if md.is_tailable() && pd.reader.is_none() {
show_error!(
"{} has appeared; following new file",
pd.display_name.quote()
);
observer.files.update_metadata(new_path, Some(md));
observer.files.update_reader(new_path)?;
_read_some = observer.files.tail_file(new_path, settings.verbose)?;
observer
.watcher_rx
.as_mut()
.unwrap()
.watch_with_parent(new_path)?;
}
}
}
}
// With -f, sleep for approximately N seconds (default 1.0) between iterations;
// We wake up if Notify sends an Event or if we wait more than `sleep_sec`.
let rx_result = observer
.watcher_rx
.as_mut()
.unwrap()
.receiver
.recv_timeout(settings.sleep_sec);
if rx_result.is_ok() {
_event_counter += 1;
_timeout_counter = 0;
}
let mut paths = vec![]; // Paths worth checking for new content to print
match rx_result {
Ok(Ok(event)) => {
if let Some(event_path) = event.paths.first() {
if observer.files.contains_key(event_path) {
// Handle Event if it is about a path that we are monitoring
paths = observer.handle_event(&event, settings)?;
}
}
}
Ok(Err(notify::Error {
kind: notify::ErrorKind::Io(ref e),
paths,
})) if e.kind() == std::io::ErrorKind::NotFound => {
if let Some(event_path) = paths.first() {
if observer.files.contains_key(event_path) {
let _ = observer
.watcher_rx
.as_mut()
.unwrap()
.watcher
.unwatch(event_path);
}
}
}
Ok(Err(notify::Error {
kind: notify::ErrorKind::MaxFilesWatch,
..
})) => {
return Err(USimpleError::new(
1,
format!("{} resources exhausted", text::BACKEND),
))
}
Ok(Err(e)) => return Err(USimpleError::new(1, format!("NotifyError: {e}"))),
Err(mpsc::RecvTimeoutError::Timeout) => {
_timeout_counter += 1;
}
Err(e) => return Err(USimpleError::new(1, format!("RecvTimeoutError: {e}"))),
}
if observer.use_polling && settings.follow.is_some() {
// Consider all files to potentially have new content.
// This is a workaround because `Notify::PollWatcher`
// does not recognize the "renaming" of files.
paths = observer.files.keys().cloned().collect::<Vec<_>>();
}
// main print loop
for path in &paths {
_read_some = observer.files.tail_file(path, settings.verbose)?;
}
if _timeout_counter == settings.max_unchanged_stats {
/*
TODO: [2021-10; jhscheer] implement timeout_counter for each file.
‘--max-unchanged-stats=n’
When tailing a file by name, if there have been n (default n=5) consecutive iterations
for which the file has not changed, then open/fstat the file to determine if that file
name is still associated with the same device/inode-number pair as before. When
following a log file that is rotated, this is approximately the number of seconds
between when tail prints the last pre-rotation lines and when it prints the lines that
have accumulated in the new log file. This option is meaningful only when polling
(i.e., without inotify) and when following by name.
*/
}
}
Ok(())
}